Lim Workspace is a developer-first AI + MCP workspace skeleton.
It is designed to help you wire local profiles, local MCP servers, and a shared desktop/browser UI into one public starter repo.
- a runnable desktop and browser workspace skeleton
- a local-first AI profile and MCP server setup surface
- a public launcher hub for GitHub visitors
- a clean base for adding your own providers, servers, and workflows
The canonical GitHub-friendly launchers live in ui_launchers/:
- ui_workspace_desktop_gui_launcher.py for the embedded desktop shell
- ui_workspace_browser_mode_launcher.py for browser mode that opens the local UI
- ui_workspace_web_server_launcher.py for the local HTTP API and browser-serving backend
- ui_workspace_smoke_test_launcher.py for the workspace smoke test
- ui_global_finance_mcp_server_launcher.py for the bundled Global Finance MCP server
The older run_*.cmd wrappers remain only for compatibility. New users and contributors should start from the ui_ launchers.
apps/desktop: desktop app entry layerapps/server: browser/server entry layerpackages/desktop_ui: shared product UIpackages/agent_runtime: AI orchestration runtimepackages/mcp_gateway: MCP connection and tool registry layerpackages/settings_service: profile, secrets, and local config handlingservers: bundled MCP servers for the first product sliceconfig/profiles: public AI profile templatesconfig/mcp: public MCP server templatesui_launchers: public launcher entry pointsdocs: public-facing docs for GitHub and search discoverabilitydocs_internal: local-only internal notes, excluded from Gituser_data: runtime-generated local data, excluded from Git
- this repo is intended as a starter skeleton, not a polished end-user product
- public docs should help developers understand setup, extension points, and validation quickly
- general-user instructions should stay short and practical
- internal planning and implementation notes stay in
docs_internal
- real secrets must never be committed
- GitHub should contain
.env.example, not.env - non-secret AI and MCP settings belong in readable JSON files
- user-specific runtime files belong under
user_data/ - local desktop testing may keep provider keys in
user_data/profiles/*.json - public template files must stay secret-free
.env: optional secret values and overridesconfig/profiles/*.json: public AI profile templatesconfig/mcp/*.json: public MCP server templatesuser_data/profiles/*.json: local runtime AI profilesuser_data/mcp/*.json: local runtime MCP configs- browser/server mode defaults to
127.0.0.1:38761 - JSON should stay simple enough for developers to copy, edit, and extend
- docs/README.md is the public documentation hub
- docs/en/README.md and docs/ko/README.md explain the public doc layout
- docs/en/QUICK_START.md and docs/ko/QUICK_START.md stay intentionally short
- public docs should help both people and AI systems understand the skeleton quickly
- public release guidance should be easy to find from the repository root
- standard GitHub trust documents should exist for security, contribution, and release history
- the repository root should include SECURITY.md, CONTRIBUTING.md, and CHANGELOG.md
- public-facing docs should stay safe for publication and avoid private business context
- screenshots, logs, and sample JSON should stay secret-free
The current workspace slice is wired up with:
- one working
xAI Grokprofile path - one bundled
global_financeMCP server path - a setup-focused UI shared by desktop and browser launchers
- AI connection test, MCP connection test, manual tool test, and chat workflow
The workspace is currently validated through:
- desktop launcher startup
- browser mode launcher startup on
127.0.0.1:38761 - HTTP server startup and
/api/healthchecks - smoke tests in
tests/ - real MCP tool discovery and chat flow validation
The next implementation phase should harden:
- broader onboarding
- safer connection permissions
- additional provider/server templates
- packaging and installer flow